Video Input

import zrt
from zrt import Agent, Pipeline
from zrt.plugins import OpenAIRealtime, OpenAIRealtimeConfig
from zrt.inference import AICousticsDenoise

AGENT_ID = "vision-agent"

pipeline = Pipeline(
    llm=OpenAIRealtime(
        config=OpenAIRealtimeConfig(model="gpt-4o-realtime-preview", voice="alloy")
    ),
    denoise=AICousticsDenoise(model_id="quail-vf-2.2-l-16khz"),
)

class VisionAgent(Agent):
    def __init__(self):
        super().__init__(
            agent_id=AGENT_ID,
            instructions="You can see the participant's video. Describe what you observe.",
            pipeline=pipeline,
        )

if __name__ == "__main__":
    # Pass the class itself (not an instance): serve() builds a fresh VisionAgent +
    # pipeline per call, which is required for correct per-call state under concurrent calls.
    zrt.serve(
        VisionAgent,
        on_ready=lambda: zrt.invoke(
            AGENT_ID, room=zrt.Room(playground=True, vision=True)
        ),
    )